Improve Performance with Nested CDATA #243
Comments
naitoh
added a commit
to naitoh/rexml
that referenced
this issue
Mar 1, 2025
GitHub: fix rubyGH-243 ## Benchmark 1 ``` $ benchmark-driver benchmark/parse_cdata.yaml Calculating ------------------------------------- before after before(YJIT) after(YJIT) dom 99.008 192.055 96.030 177.653 i/s - 100.000 times in 1.010020s 0.520685s 1.041338s 0.562894s sax 101.858 205.223 99.488 194.657 i/s - 100.000 times in 0.981757s 0.487274s 1.005148s 0.513723s pull 99.614 205.030 100.284 196.212 i/s - 100.000 times in 1.003879s 0.487733s 0.997173s 0.509654s stream 102.506 205.073 100.324 195.893 i/s - 100.000 times in 0.975549s 0.487632s 0.996772s 0.510483s Comparison: dom after: 192.1 i/s after(YJIT): 177.7 i/s - 1.08x slower before: 99.0 i/s - 1.94x slower before(YJIT): 96.0 i/s - 2.00x slower sax after: 205.2 i/s after(YJIT): 194.7 i/s - 1.05x slower before: 101.9 i/s - 2.01x slower before(YJIT): 99.5 i/s - 2.06x slower pull after: 205.0 i/s after(YJIT): 196.2 i/s - 1.04x slower before(YJIT): 100.3 i/s - 2.04x slower before: 99.6 i/s - 2.06x slower stream after: 205.1 i/s after(YJIT): 195.9 i/s - 1.05x slower before: 102.5 i/s - 2.00x slower before(YJIT): 100.3 i/s - 2.04x slower ``` - YJIT=ON : 1.84x - 1.95x faster - YJIT=OFF : 1.94x - 2.04x faster ### Benchmark Code - benchmark/parse_cdata.yaml ```ruby loop_count: 100 contexts: - name: before gems: rexml: 3.4.1 require: false prelude: require 'rexml' - name: after prelude: | $LOAD_PATH.unshift(File.expand_path("lib")) require 'rexml' - name: before(YJIT) gems: rexml: 3.4.1 require: false prelude: | require 'rexml' RubyVM::YJIT.enable - name: after(YJIT) prelude: | $LOAD_PATH.unshift(File.expand_path("lib")) require 'rexml' RubyVM::YJIT.enable prelude: | require 'rexml/document' require 'rexml/parsers/sax2parser' require 'rexml/parsers/pullparser' require 'rexml/parsers/streamparser' require 'rexml/streamlistener' def build_xml(n_depth) xml = "<?xml version=\"1.0\"?>\n" + "<root>Test</root>\n" + "<!" + "[CDATA[" * n_depth + "]]>\n" end xml = build_xml(100000) class Listener include REXML::StreamListener end benchmark: 'dom' : REXML::Document.new(xml) 'sax' : REXML::Parsers::SAX2Parser.new(xml).parse 'pull' : | parser = REXML::Parsers::PullParser.new(xml) while parser.has_next? parser.pull end 'stream' : REXML::Parsers::StreamParser.new(xml, Listener.new).parse ``` ## Benchmark 2 - depth 140000001 : 1.91x faster - befofe ``` $ ruby nested_cdata.rb Trying depth 1 Elapsed time: 0.000148 seconds Trying depth 10000001 Elapsed time: 1.069345 seconds Trying depth 20000001 Elapsed time: 2.08041 seconds Trying depth 30000001 Elapsed time: 3.120498 seconds Trying depth 40000001 Elapsed time: 4.169959 seconds Trying depth 50000001 Elapsed time: 5.220279 seconds Trying depth 60000001 Elapsed time: 6.343796 seconds Trying depth 70000001 Elapsed time: 7.336923 seconds Trying depth 80000001 Elapsed time: 8.357979 seconds Trying depth 90000001 Elapsed time: 9.438741 seconds Trying depth 100000001 Elapsed time: 10.564045 seconds Trying depth 110000001 Elapsed time: 11.505168 seconds Trying depth 120000001 Elapsed time: 12.696258 seconds Trying depth 130000001 Elapsed time: 13.81369 seconds Trying depth 140000001 Elapsed time: 14.772476 seconds Trying depth 150000001 ^CError at depth 150000001: ``` - after ``` $ ruby nested_cdata.rb Trying depth 1 Elapsed time: 0.000135 seconds Trying depth 10000001 Elapsed time: 0.546576 seconds Trying depth 20000001 Elapsed time: 1.110255 seconds Trying depth 30000001 Elapsed time: 1.592627 seconds Trying depth 40000001 Elapsed time: 2.09499 seconds Trying depth 50000001 Elapsed time: 2.700936 seconds Trying depth 60000001 Elapsed time: 3.150459 seconds Trying depth 70000001 Elapsed time: 3.665467 seconds Trying depth 80000001 Elapsed time: 4.168667 seconds Trying depth 90000001 Elapsed time: 4.999614 seconds Trying depth 100000001 Elapsed time: 5.582072 seconds Trying depth 110000001 Elapsed time: 5.971595 seconds Trying depth 120000001 Elapsed time: 6.423684 seconds Trying depth 130000001 Elapsed time: 7.108247 seconds Trying depth 140000001 Elapsed time: 7.714127 seconds Trying depth 150000001 ^CError at depth 150000001: ``` ### Benchmark Code - nested_cdata.rb ```ruby require "rexml/document" include REXML (1..1000000000).step(10000000) do |depth| puts "Trying depth #{depth}" string = "<?xml version=\"1.0\"?>\n" + "<root>Test</root>\n" + "<!" + "[CDATA[" * depth + "]]>\n" start = Time.now begin doc = Document.new(string) rescue Exception => e puts "Error at depth #{depth}: #{e}" break end elapsed_time = Time.now - start puts "Elapsed time: #{elapsed_time} seconds" end ```
|
I don't think it is a problem. Large data takes time. Not related to nested cdata. size = 1000000
nested_cdata = "<?xml version=\"1.0\"?><root>Test</root><!" + "[CDATA[" * size + "]]>"
normal_cdata = "<?xml version=\"1.0\"?><root>Test</root><![CDATA[" + 'aaaaaaa' * size + "]]>"
normal_text = "<?xml version=\"1.0\"?><root>Test" + 'aaaaaaa' * size + "</root>"
normal_tags = "<?xml version=\"1.0\"?><root>Test" + '<b></b>' * size + "</root>"
# Size are almost the same.
[nested_cdata, normal_cdata, normal_text, normal_tags].map(&:size)
# => [7000043, 7000050, 7000038, 7000038]
REXML::Document.new(nested_cdata);
# processing time: 0.130978s
REXML::Document.new(normal_cdata);
# processing time: 0.109913s
REXML::Document.new(normal_text);
# processing time: 0.144391s
REXML::Document.new(normal_tags);
# processing time: 5.863289sThe slowness is just caused by the input size. Time consumption of parsing nested cdata, non-nested normal cdata, normal text node is almost the same. |
naitoh
added a commit
to naitoh/rexml
that referenced
this issue
Mar 2, 2025
## Why? See: ruby#243 ## Benchmark (Comparison with rexml 3.4.1) ``` $ benchmark-driver benchmark/parse_cdata.yaml Calculating ------------------------------------- rexml 3.4.1 master 3.4.1(YJIT) master(YJIT) dom 648.361 1.178k 591.590 1.046k i/s - 100.000 times in 0.154235s 0.084913s 0.169036s 0.095627s sax 699.061 1.378k 651.148 1.196k i/s - 100.000 times in 0.143049s 0.072549s 0.153575s 0.083611s pull 699.271 1.379k 660.275 1.210k i/s - 100.000 times in 0.143006s 0.072527s 0.151452s 0.082622s stream 701.725 1.383k 659.483 1.228k i/s - 100.000 times in 0.142506s 0.072307s 0.151634s 0.081455s Comparison: dom master: 1177.7 i/s master(YJIT): 1045.7 i/s - 1.13x slower rexml 3.4.1: 648.4 i/s - 1.82x slower 3.4.1(YJIT): 591.6 i/s - 1.99x slower sax master: 1378.4 i/s master(YJIT): 1196.0 i/s - 1.15x slower rexml 3.4.1: 699.1 i/s - 1.97x slower 3.4.1(YJIT): 651.1 i/s - 2.12x slower pull master: 1378.8 i/s master(YJIT): 1210.3 i/s - 1.14x slower rexml 3.4.1: 699.3 i/s - 1.97x slower 3.4.1(YJIT): 660.3 i/s - 2.09x slower stream master: 1383.0 i/s master(YJIT): 1227.7 i/s - 1.13x slower rexml 3.4.1: 701.7 i/s - 1.97x slower 3.4.1(YJIT): 659.5 i/s - 2.10x slower ``` - YJIT=ON : 1.76x - 1.83x faster - YJIT=OFF : 1.82x - 1.97x faster
naitoh
added a commit
to naitoh/rexml
that referenced
this issue
Mar 2, 2025
## Why? GitHub: fix ruby#243 ## Benchmark (Comparison with rexml 3.4.1) ``` $ benchmark-driver benchmark/parse_cdata.yaml Calculating ------------------------------------- rexml 3.4.1 master 3.4.1(YJIT) master(YJIT) dom 648.361 1.178k 591.590 1.046k i/s - 100.000 times in 0.154235s 0.084913s 0.169036s 0.095627s sax 699.061 1.378k 651.148 1.196k i/s - 100.000 times in 0.143049s 0.072549s 0.153575s 0.083611s pull 699.271 1.379k 660.275 1.210k i/s - 100.000 times in 0.143006s 0.072527s 0.151452s 0.082622s stream 701.725 1.383k 659.483 1.228k i/s - 100.000 times in 0.142506s 0.072307s 0.151634s 0.081455s Comparison: dom master: 1177.7 i/s master(YJIT): 1045.7 i/s - 1.13x slower rexml 3.4.1: 648.4 i/s - 1.82x slower 3.4.1(YJIT): 591.6 i/s - 1.99x slower sax master: 1378.4 i/s master(YJIT): 1196.0 i/s - 1.15x slower rexml 3.4.1: 699.1 i/s - 1.97x slower 3.4.1(YJIT): 651.1 i/s - 2.12x slower pull master: 1378.8 i/s master(YJIT): 1210.3 i/s - 1.14x slower rexml 3.4.1: 699.3 i/s - 1.97x slower 3.4.1(YJIT): 660.3 i/s - 2.09x slower stream master: 1383.0 i/s master(YJIT): 1227.7 i/s - 1.13x slower rexml 3.4.1: 701.7 i/s - 1.97x slower 3.4.1(YJIT): 659.5 i/s - 2.10x slower ``` - YJIT=ON : 1.76x - 1.83x faster - YJIT=OFF : 1.82x - 1.97x faster Co-authored-by: Sutou Kouhei <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
I think we should improve the implementation to handle nested CDATA more efficiently.
Result:
The text was updated successfully, but these errors were encountered: